Decoding letter position in word reading

Abstract

A fundamental computation underlying visual word recognition is the ability to transform a set of letters into a visual word form. Neuropsychological data suggest that letter position within a word may be independently affected by brain damage, resulting in a dissociable subtype of peripheral dyslexia. Here we used functional magnetic resonance imaging and supervised machine learning techniques to classify letter position based on activation patterns evoked during reading Hebrew words. Across the entire brain, activity patterns in the left intraparietal sulcus provided the best classification accuracy (80%) with respect to letter position. Importantly, the same set of voxels that showed highest classification performance of letter position using one letter-of-interest also showed highest classification performance using a different letter-of-interest. A functional connectivity analysis revealed that activity in these voxels co-varied with activity in the Visual Word Form Area, confirming cross-talk between these regions during covert reading. The results converge with reports of patients with acquired letter position dyslexia, who suffer from left occipito-parietal lesions. These findings provide direct and novel evidence for the role of left IPS within the reading network in processing relative letter positions.

Introduction

The ability of the human visual system to conduct efficient and successful orthographic analysis of written words is a fundamental step in reading. Despite comprehensive research, the functional and structural properties of the neural mechanisms underlying successful visual word recognition are not yet fully elucidated. During the last two decades, advances in neuroimaging measurements provided consistent evidence that visual word processing relies on both ventral and dorsal visual networks. Within the ventral pathway, a series of regions including the Visual Word Form Area (VWFA, located in the posterior occipitotemporal sulcus), were found to respond robustly and consistently to written word stimuli (Cohen et al., 2002, Price and Devlin, 2011, Wandell, 2011). Within the dorsal pathway, regions in the left and right posterior intraparietal sulcus (IPS) and the adjacent temporoparietal junction were found to be engaged during reading tasks (Cohen et al., 2008, Gabrieli, 2009, Reilhac et al., 2012).

Computational modeling (Gomez et al., 2008, McClelland and Rumelhart, 1981), cognitive and neuropsychological studies (Coltheart, 1981, Ellis et al., 1987, Friedmann and Gvion, 2001, Rastle, 2007) argue that encoding of letter position within words may be selectively impaired. Other computational models, supported by neuropsychological studies showing non selective letter-order deficits, suggest that orthographic analysis units do not contain precise information about letter position (Grainger and Van Heuven, 2003, Katz and Sevush, 1989, Whitney, 2001).

Several studies have established a causal role for the VWFA in orthographic processing (Cohen et al., 2003, Dehaene, 2009, Gaillard et al., 2006, Rauschecker et al., 2011, Rauschecker et al., 2012). The VWFA is argued to respond to words as holistic constructs (Glezer, Jiang, & Riesenhuber, 2009) suggesting that encoding of letter position might be performed elsewhere. On the other hand, the VWFA could be indirectly involved in encoding letter positions, as part of a ventral cascade of areas that are sensitive to larger and larger letter combinations (Dehaene et al., 2005, Thesen et al., 2012, Vinckier et al., 2007). We hypothesized that dorsal parietal areas are natural candidates for encoding letter positions. This hypothesis builds on their well-documented involvement in spatial attention (Shafritz et al., 2002, Xu, 2007), as well as their known involvement in visual word processing, particularly of spatially distorted words (Cohen et al., 2008, Vinckier et al., 2006). Further support for this hypothesis is provided by evidence from two documented patients who sustained left occipital-parietal lesions, displaying acquired letter position dyslexia (LPD) – a selective impairment in encoding letter positions (Friedmann & Gvion, 2001).

Here, we used fMRI measurements and multi-voxel pattern analysis (MVPA) techniques in a searchlight strategy to examine which cortical regions provide the most information about the position of a letter-of-interest within visually presented words. Using this data driven approach across the entire brain, we show that the left intraparietal sulcus (L-IPS) provides the best classification with respect to letter position, regardless of letter identity. These findings provide clear support for the hypothesis that letter position encoding in word reading can be supported by activity patterns of neuronal populations in the left parietal cortex.